TaughtHuman decisions for better models

A model inference API that names exactly what ran.

Taught’s canonical serving contract ties each request to a promoted endpoint revision, exact model identity, input hash, predictions, pricing rule, and usage receipt.

Request early access Read the API contract
Evidence registerA prediction should arrive with its operational evidence.

The response identifies the endpoint and model version instead of returning anonymous output from a floating latest model.

  1. Authenticate
    Scoped bearer key

    Server-side clients call an endpoint within project authority.

  2. Protect
    Idempotent request

    A required key makes retries safe for the same bytes and parameters.

  3. Predict
    Versioned model output

    Normalized detections identify the endpoint revision and checkpoint.

  4. Receipt
    Metered evidence

    Input identity, duration, pricing, usage, and outcome remain together.

Operational model serving without a floating identity.

Applications need more than prediction arrays. They need retry safety, stable endpoint authority, exact model provenance, and accountable usage. The Taught v1 contract puts those concerns in the request and response.

Endpoint revision
The request resolves through a named, promoted endpoint configuration.
Exact model identity
Provider, architecture, taxonomy version, weight hash, and class-map hash identify what produced the output.
Idempotency record
Retries with the same key are safe; conflicting reuse is rejected rather than double-counted.
Usage receipt
Measured duration, accelerator evidence, credits, pricing version, and outcome are returned together.

From promotion authority to client receipt

  1. 01
    Promote a compatible model

    A project gate changes the active version before an endpoint can serve it.

  2. 02
    Create a versioned endpoint

    The endpoint revision fixes model, taxonomy, pricing, and serving constraints.

  3. 03
    Call from a trusted server

    Use the deployment’s supplied service origin, bearer key, and idempotency key.

  4. 04
    Retain the receipt

    Store the exact model and usage evidence beside the application outcome.

Serving availability is deployment-gated

  • The documentation defines the canonical v1 inference contract; it does not assert a public API hostname.
  • Hosted inference must be reported available by the target deployment before clients call it.
  • MCP and A2A are projections over the same authority and remain separately deployment-gated.

Questions teams ask before they begin

Direct answers based on the current Taught product and deployment boundaries.

What URL should an application call?

Use the service origin supplied for the target deployment. The marketing domain is not presented as a verified public API hostname.

Why is an Idempotency-Key required?

It lets a client retry the same request safely and lets the service reject reuse with different bytes or parameters.

Does the response identify the model version?

Yes. The canonical response includes the endpoint revision and exact model evidence, including hashes for the weights and class map.

Bring the decisions, data, and model version into one record.

Request access for a labeling project, a custom-model workflow, or an integration with a promoted model.

Request early access